画像クレジット: Lottie animation by Centre Robotics (LottieFiles Free, used with credit). · source
I spent yesterday morning scrolling through OpenAI's latest announcement about reaching one million business customers, and honestly, my first reaction was skepticism. A million businesses sounds massive. It sounds like AI has officially gone mainstream in the enterprise. But the more I read, the more I found myself asking: what exactly are we counting here?
Let me back up. OpenAI published a celebratory post highlighting companies like PayPal, Virgin Atlantic, BBVA, Cisco, Moderna, and Canva as examples of organizations "transforming the way work gets done with AI." These are real companies doing real things. Canva's using it for design workflows. Moderna's apparently integrating it into research processes. That's genuinely interesting.
But here's where I get stuck. The announcement doesn't distinguish between a 50,000-person enterprise with deep API integrations and a five-person startup that signed up for ChatGPT Teams last month. Both count as "business customers." Both contribute to that one million figure. I initially thought this was a minor quibble, but after reading through their various case studies, I think it actually matters a lot for understanding where AI adoption really stands.
The case studies tell a more specific story. Take Choco, a food distribution company that's using OpenAI's APIs to automate parts of their supply chain. According to OpenAI's blog, they've built AI agents that streamline ordering processes between restaurants and suppliers. This is the kind of implementation that actually changes how a business operates, not just adding a chatbot to a website.
Then there's OpenAI itself, which (somewhat recursively) published details about using its own tools to handle customer support and inbound sales. Their support team apparently cut response times significantly by having AI draft initial responses that human agents then review. For sales, they built an assistant that delivers personalized answers to leads at scale.
関連記事
More in AI Models
The company's new 'Agentic Commerce Protocol' sounds impressive, but I've seen enough automation hype cycles to know the difference between demos and deployment.
Robert "Bob" Macintosh · 56 mins ago · 4 min
The company just dropped four papers on watching AI think out loud. It's genuinely interesting work, but let's not pretend we've solved alignment.
Mark Kowalski · 56 mins ago · 6 min
GPT-5.4 mini and nano aren't about chatbots. They're about running inference on edge hardware without melting your power budget.
James Chen · 56 mins ago · 4 min
The company says it built safety 'at the foundation.' I have questions.
You might be wondering why I'm spending so much time on these individual examples. Here's why: the gap between "we have ChatGPT licenses" and "we've rebuilt workflows around AI" is enormous. And the one million number doesn't tell us anything about that distribution.
What we don't know is significant. OpenAI hasn't disclosed how many of these customers are on paid enterprise tiers versus smaller team plans. They haven't said what percentage are actively using APIs versus just the chat interface. They haven't shared retention numbers, or how usage changes after the initial onboarding period. These aren't unreasonable things to want to know when evaluating a milestone like this.
I should be clear: I'm not saying the number is meaningless. A million is a lot of organizations, and even if most are small, that represents real money and real adoption. But tbh, I've seen too many AI announcements that sound transformative until you look at the specifics.
The enterprise examples are more compelling. When OpenAI highlights that BBVA (a major Spanish bank) or Cisco are customers, that's different from counting startups. Financial services companies have compliance requirements, security reviews, legal approvals. Getting AI tools through that gauntlet means something. Healthcare and life sciences companies like Moderna face even more scrutiny. If they're genuinely integrating these tools into research workflows, that's a real signal about where the technology stands.
But again, we don't have numbers on how many of the million fall into this category versus the "signed up for a team plan and uses it occasionally" bucket.
Here's what I think is actually happening. AI tools have reached the "spreadsheet" phase of adoption. Basically everyone has access, most people use them sometimes, and a small percentage have figured out how to make them genuinely transformative for their work. The one million number captures all of that, which is why it feels both impressive and underwhelming at the same time.
The Choco example is instructive here. They didn't just add ChatGPT to their workflow. They built custom AI agents using the API, integrated them into their existing systems, and fundamentally changed how orders flow between restaurants and suppliers. That requires engineering resources, clear use cases, and organizational buy-in. Most companies aren't there yet.
I think OpenAI knows this, which is why they're publishing detailed case studies alongside the headline number. The case studies do the work of showing what's possible when you go deep. The million number does the work of showing that adoption is broad. Both are true. Neither tells the complete story.
The support and sales examples are revealing in a different way. OpenAI using its own tools for customer support is interesting because it shows the current state of AI-assisted work. The support team uses AI to draft responses, but humans still review and send them. The sales assistant provides personalized answers, but presumably humans still close deals. This is augmentation, not automation. It's useful, it probably saves time, but it's not the "AI does the job" future that some headlines suggest.
And honestly, I'm not sure that's a bad thing. The augmentation model, where AI handles the first draft or the initial research or the routine responses, seems more sustainable than full automation for most knowledge work. It's just less exciting than the hype suggests.
What would make me take this milestone more seriously? A few things. Breakdown by company size. Retention rates after six months. Average revenue per customer. Usage metrics (API calls, messages sent, whatever). How many customers have built custom integrations versus using off-the-shelf products. Any of these would help distinguish between "AI is everywhere" and "AI is everywhere but mostly in the background."
I reached out to OpenAI to ask about some of these details but haven't heard back yet. If they respond, I'll update this piece.
For now, here's my read. One million business customers is a real milestone. It means OpenAI has built distribution that reaches across industries, company sizes, and use cases. The enterprise examples (Cisco, BBVA, Moderna) suggest the technology is mature enough for serious organizations with serious requirements. The case studies show what's possible when companies invest in real integration.
But the headline number alone doesn't tell us how deep AI adoption has actually gone. It's too early to say whether most of these million customers are fundamentally changing how they work, or just experimenting. The answer is probably "some of both," but the ratio matters.
I think we're still in the early innings of figuring out what AI actually does for businesses. The million customer milestone is a marker along that path, not the destination. And I'll be more interested in the two million announcement if it comes with more detail about what those customers are actually doing.